Skip to main content
BIO·ATLASby Bayes Pharma.ai
Search
AccountStart evaluation
AI drug discovery models/Small-Molecule & Chemistry/Uni-Mol / Uni-Mol2
model-family passport · Review date not recorded

Uni-Mol / Uni-Mol2

3D molecular foundation models trained on atoms, topology and conformational geometry.

3/7Evidence fields documented
60-SECOND EVALUATION VIEW

What should a scientist know before using Uni-Mol / Uni-Mol2?

SupportedEvidence supports the stated context with explicit boundaries
Best suited forRepresentation · Generation
Evidence supportsMolecular representation/property benchmarks: Open evaluation
Evidence does not establishUniversal superiority, therapeutic success, clinical utility or regulatory acceptance.
Major limitationPerformance depends on the evaluation dataset and operating conditions.
CompareAdd to evaluationView evidence
Current registry recordVersion history not yet curated1 recorded release · Review date not recorded. A newer version is not assumed to be universally better.

What it is

Uni-Mol and Uni-Mol2 learn transferable representations from 3D molecular conformers for chemistry prediction and structure-aware tasks.

Evidence trail

BioAtlas keeps the path from source to decision visible. A connection records provenance; it does not imply that evidence is sufficient for every context.

Sources3 connectedPrimary resources and normalized claims
Claims1 normalizedProtein–ligand pose prediction
EntityUni-Mol / Uni-Mol2model-family · Version history not yet curated
ReviewReview date not recordedReview date not claimed
ConclusionContext requiredAdd to an evaluation before operational use

Model passport

Entity typemodel-family
OrganizationDP Technology / DeepModeling
Model family introducedNot normalized
AccessOpen source
Commercial useAllowed / verify checkpoint terms
DeploymentSelf-hosted
ComputeGPU recommended
Domainschemistry
Biology → representation → computation → evidence

How Uni-Mol / Uni-Mol2 represents biology

model-familychemistry

Category is navigation. These fields describe the model-specific computational transformation and deliberately override broad category defaults.

1 · Biological inputs
Molecular graph3D conformer
2 · Input representation
Atom tokensBond/topology graphPairwise geometry
3 · Internal representation
Atom embeddingsPair embeddingsMolecular embedding
4 · Architecture
Two-track molecular Transformer
5 · Learning objective
Self-supervised 3D molecular pretraining
6 · Output representation
Dense vectorsScalar/task scores

Biological scale

molecule

Modalities & tasks

MoleculeRepresentationGeneration

Registry, claims and frontier intelligence

Versioned registry

Version history not yet curated

1 version record · release year not yet normalized. Model-family identity remains separate from capability and access changes.

Explore version lineage →
Benchmark claim ledger

1 normalized claim

Protein–ligand pose prediction · Molecular representation/property benchmarks

Open claim intelligence →

Inputs and outputs

Inputs

Molecular graph3D conformer

Outputs

Molecular embeddingsProperty predictions

Scientific and technical profile

Scientific principles

3D molecular representation learningGeometric deep learning

Technology

Molecular TransformerPair representation
Ideas before algorithms

Scientific lineage

Explore all foundations

These are transparent concept matches—not claims that one scientist alone caused this model. Each connection is based on the model’s recorded domain, scientific principles, technical terms or an explicit lineage link.

Computational intelligence

Transformer self-attention

Ashish Vaswani and colleagues

Protein, genome, molecule and single-cell foundation models use attention to learn dependencies across biological sequences and multimodal inputs.

Matched concepts: transformer, foundation model
Physical chemistry

Statistical mechanics and the Boltzmann distribution

Ludwig Boltzmann

Conformational ensembles, molecular simulations, temperature scaling, sampling and energy-based generative models rely on this statistical view.

Matched concepts: conformation
Molecular recognition

Induced-fit binding

Daniel E. Koshland Jr.

Flexible docking, conformational selection, protein motion and ligand-induced pocket changes are modern extensions of this idea.

Matched concepts: conformation
Medicinal chemistry & pharmacology

Quantitative structure–activity relationships

Corwin Hansch

Classical QSAR established the central premise that molecular features can predict potency and guide optimization—the conceptual ancestor of modern molecular machine learning.

Matched concepts: property prediction
Molecular recognition

Concerted allostery

Jacques Monod, Jeffries Wyman & Jean-Pierre Changeux

Allosteric drug design exploits remote pockets to modulate function, selectivity and resistance without competing at the active site.

Matched concepts: conformational

Evaluation evidence

Dataset or evaluationMolecular representation/property benchmarks
Task or metric3D molecular representation
Evidence statusOpen evaluation
Open source ↗

Task-specific evidence only; not comparable as a universal leaderboard score.

Protein–ligand pose prediction

Molecular representation/property benchmarks

Version history not yet curated · Split details not yet normalized
developer-reported

A structured benchmark claim is recorded; consult the linked source for numeric values and protocol details.

Claim caveats
  • Protocol, split and implementation details must match before comparing this claim with another result.
Open benchmark claim →

Known limitations

  • Performance depends on the evaluation dataset and operating conditions.
  • Task-specific benchmark results should not be compared across unlike domains.
  • Outputs require task-specific scientific and experimental validation.

Milestones

Not normalized

Uni-Mol2 scales the family to large 3D molecular pretraining corpora.